As the artificial intelligence - the main trends and obstacles is improved

Development of technologies of collecting and processing of large volumes of information along with successful researches in the field of algorithms machine learning gave the amazing growth of technologies which can be identified with AI. Achievements of ultraprecise neural networks in the solution of a problem of classification and image identification on the basis of the ImageNet database in 2012, a victory AlphaGo in March, 2016 over the person in a Go game, achievement by the companies Microsoft and IBM the accuracy of recognition of the telephone speech comparable to the person in 2017 – all this caused splash in interest in artificial intelligence.

Are laid great hopes on artificial intelligence

The number of participants of the last NIPS conferences devoted to machine learning exceeded 5,000 people and was almost equal to the number of conferees of AAAI (Association for the Advancement of Artificial Intelligence) to second "winter" of AI. Interest in algorithms and the AI systems went beyond the academic conferences and entailed investment activity in the industry. Optimistically configured visionaries consider that AI is capable to solve all most critical problems of mankind. More careful researchers analyze technical issues of reproduction of "work of a brain" in the next decades and argue on moral ethical aspect of implementing solutions on the basis of AI.

For someone AI became a synonym of robots, for someone - neural networks and deep learning, someone is inclined to identify it with a plant. Some experts demonize artificial intelligence, predicting creation in several decades of such AI which will exceed the intellectual level of the person without having at the same time moral ethical standards; other researchers lay on AI hopes for the solution of all problems of mankind, including aging, death and natural disasters.

Algorithms

The modern systems of artificial intelligence can beat the grand master on chess, but the intelligent system which could "beat" the child in the "developing" games is not created yet.

Intelligent systems which became an integral part of modern reality is not "true" AI. Modern pattern recognition systems, speeches, a natural language and forecasting are based on recognition of patterns in data bulks. The AI numerous systems from the most powerful, such as DeepBlue or Watson, to the simplest algorithms of credit scoring and spam filtering or fraudulent financial transactions are capable to solve effectively only a set of the same problems and have no flexibility and variety of functions of intelligence even of the two-year-old child. These systems cannot explain and understand the world around, they have no imagination, they are weak in questions of strategic planning and cannot create new models of reality. A basis of all modern intelligent systems - training at large volumes of data which is not similar to the training of the person to something new based on understanding.

Computational Cognitive Science laboratories of Stanford University, the Massachusetts Institute of Technology (MIT), the University of California in Berkeley and many other largest universities of the world are engaged in studying and modeling of computing bases of learning process and conclusions. This direction is one of developing and the most demanded, for example, in the field of unmanned vehicles.

At the same time, to already existing approaches of machine learning in the solution of a number of tasks new calls are put. Often effective use of results of machine learning in practice from model requires an interpretiruyemost of results. For example, the model should not just specify the probability of this or that diagnosis in a x-ray analysis result, but also be able to specify on the basis of what section in the picture the conclusion was drawn.

Relevant is a question of preparation of qualitative data sets for training of models of machine learning. Application researches in different areas of training with a reinforcement (Reinforcement learning) as machine learning in dynamically changing environment are perspective. As the most part of the saved-up data is not marked for algorithms of machine learning, relevant are algorithms with partial involvement of the teacher (Semi-supervised learning) and also algorithms of unsupervised learning. Identification of causes and effect relationships in data bulks also is topical issue for many researchers.

Investments

Important component of development of AI is that, how fast and investments into AI made on a wave of the increased interest in the industry can effectively pay off. Specialists in capital management of Morgan Stanley note that at the current growth rates the industry can grow to $1 trillion by 2050. At the same time different solutions AI can be directed both to profit earning, and to additional business optimization.

According to the research conducted by MIT together with BCG at this stage of development of AI pioneers of the industry select implementation of those solutions which are directed to increase in profit and creation of new business models. As a rule, concern internal processes of the solution AI in the last queue as loss of foreign market and profit in the conditions of the competition is always more critical for business, than internal costs of imperfection of business processes. The companies which do not follow this principle risk during an era of rapid development of business models on the basis of AI as in the market there are always younger competitors who have no massive settled business processes requiring optimization. In effectiveness assessment from implementation of this or that AI technology the success of development and further investment into AI lies. At the same time the range of the tasks connected with optimization will be relevant for large business in the near future.

In the Russian market the active growth of investments into the AI systems is also observed from large industrial holdings, investment funds and the state. Problems which often are solved by industrial giants of the Russian market mostly are optimization, however they allow the enterprises to optimize production so that to cover the increasing sales markets that gives hope for future investments into solving of tasks, the new business models connected with creation. However the weak integration of data sources can become a serious obstacle for creation of new business models for large enterprises in the near future. Implementation of uniform data warehouses and active mutually advantageous data exchange in corporations can serve creation of new products, sales markets and business models. Thus, implementation of technology of Big Data is the key moment for success of implementation of the AI modern systems at the level of large industrial holdings and the state.

Development trends

The AI systems will hardly be able to save mankind from environmental disasters and climate change in the nearest future, but they will help to avoid technogenic catastrophes and to reduce damage from them. It is unlikely the AI systems will save mankind from financial crises, but they can improve identification of fraudulent transactions and hidden the factors influencing a financial market behavior. It is unlikely the AI systems will be able to prevent mass social riots, but they help to reduce crime rate already now and to optimize activity of law-enforcement services.

Top trends in the industry of AI are:

• The improvement and giving of nuances to already existing solutions in the market of a system of machine learning and AI which is implemented as a result of the competition and investment cycles.
• Development of the AI adaptation functions, including communicative skills and perception of human emotions. For effective addition of human activity with computing power of the AI systems further researches in the field of development of interactive functions of interaction of the AI systems with the person, increase in the explaining role of algorithms of machine learning and also studying of influence of AI on society are demanded.
• Further researches in the field of algorithms of unsupervised learning and algorithms with partial involvement of the teacher for search of new solutions in the conditions of limited data sets.
• Continuation of researches in the field of Computational Cognitive Science which in the future will help to approach creation of more flexible and the AI multifunction systems and will allow to improve modern robots.

These trends will be how dynamically implemented in the near future depends on many external factors. Whether waits for AI one more "winter" or we will become witnesses of creation of the first self-training systems who will not be needs a training on huge information volumes for accomplishment of one or several functions, depend on many external factors, in particular from political and economic stability and also from readiness of both people, and technologies to interact and solve problems jointly. However to predict, how fast it will occur, not in forces even the most perfect system of machine learning yet.

Author: Tatyana Zobnina

See Also

• Artificial intelligence (world market)
• Artificial intelligence (market of Russia)

Robotics